Final Project
Preston Scibek
May 13, 2018
Intro
url_pitching = https://baseballsavant.mlb.com/statcast_search?hfPT=&hfAB=&hfBBT=&hfPR=&hfZ=&stadium=&hfBBL=&hfNewZones=&hfGT=R%7C&hfC=&hfSea=2018%7C&hfSit=&player_type=pitcher&hfOuts=&opponent=&pitcher_throws=&batter_stands=&hfSA=&game_date_gt=&game_date_lt=&team=BOS&position=&hfRO=&home_road=&hfFlag=&metric_1=&hfInn=&min_pitches=0&min_results=0&group_by=name&sort_col=pitches&player_event_sort=h_launch_speed&sort_order=desc&min_abs=0#results
url_batting = https://baseballsavant.mlb.com/statcast_search?hfPT=&hfAB=&hfBBT=&hfPR=&hfZ=&stadium=&hfBBL=&hfNewZones=&hfGT=R%7C&hfC=&hfSea=2018%7C&hfSit=&player_type=batter&hfOuts=&opponent=&pitcher_throws=&batter_stands=&hfSA=&game_date_gt=&game_date_lt=&team=BOS&position=&hfRO=&home_road=&hfFlag=&metric_1=&hfInn=&min_pitches=0&min_results=0&group_by=name&sort_col=pitches&player_event_sort=h_launch_speed&sort_order=desc&min_abs=0#results
This data comes from the mlb statcast dataset. The data can be found at the above urls by clicking on the icon above the graphs column in the table. Then load the data as shown below.
library(tidyverse)
library(stringr)
library(rvest)
library(ggplot2)
library(plotly)
library(tree)
library(randomForest)
library(png)
library(caret)
library(rpart)
final_pitching_data_table <- read_csv("redsox_pitching_5132018.csv")
final_batting_data_table <- read_csv("redsox_batting_5132018.csv")
head(final_pitching_data_table)## # A tibble: 6 x 87
## pitch_type game_date release_speed release_pos_x release_pos_z
## <chr> <date> <chr> <chr> <chr>
## 1 FF 2018-05-12 96.6 -3.1489 4.8075
## 2 KC 2018-05-12 86.8 -3.0017 4.7448
## 3 FF 2018-05-12 96.3 -3.0122 4.8443
## 4 FF 2018-05-12 96.6 -2.8003 4.8929
## 5 FF 2018-05-12 97.1 -3.0436 4.8279
## 6 KC 2018-05-12 84.3 -2.9241 4.8362
## # ... with 82 more variables: player_name <chr>, batter <int>,
## # pitcher <int>, events <chr>, description <chr>, spin_dir <chr>,
## # spin_rate_deprecated <chr>, break_angle_deprecated <chr>,
## # break_length_deprecated <chr>, zone <chr>, des <chr>, game_type <chr>,
## # stand <chr>, p_throws <chr>, home_team <chr>, away_team <chr>,
## # type <chr>, hit_location <chr>, bb_type <chr>, balls <int>,
## # strikes <int>, game_year <int>, pfx_x <chr>, pfx_z <chr>,
## # plate_x <chr>, plate_z <chr>, on_3b <chr>, on_2b <chr>, on_1b <chr>,
## # outs_when_up <int>, inning <int>, inning_topbot <chr>, hc_x <chr>,
## # hc_y <chr>, tfs_deprecated <chr>, tfs_zulu_deprecated <chr>,
## # pos2_person_id <chr>, umpire <chr>, sv_id <chr>, vx0 <chr>, vy0 <chr>,
## # vz0 <chr>, ax <chr>, ay <chr>, az <chr>, sz_top <chr>, sz_bot <chr>,
## # hit_distance_sc <chr>, launch_speed <chr>, launch_angle <chr>,
## # effective_speed <chr>, release_spin_rate <chr>,
## # release_extension <chr>, game_pk <int>, pos1_person_id <chr>,
## # pos2_person_id_1 <chr>, pos3_person_id <chr>, pos4_person_id <chr>,
## # pos5_person_id <chr>, pos6_person_id <chr>, pos7_person_id <chr>,
## # pos8_person_id <chr>, pos9_person_id <chr>, release_pos_y <chr>,
## # estimated_ba_using_speedangle <chr>,
## # estimated_woba_using_speedangle <chr>, woba_value <chr>,
## # woba_denom <chr>, babip_value <chr>, iso_value <chr>,
## # launch_speed_angle <chr>, at_bat_number <int>, pitch_number <int>,
## # pitch_name <chr>, home_score <int>, away_score <int>, bat_score <int>,
## # fld_score <int>, post_away_score <int>, post_home_score <int>,
## # post_bat_score <int>, post_fld_score <int>head(final_batting_data_table)## # A tibble: 6 x 87
## pitch_type game_date release_speed release_pos_x release_pos_z
## <chr> <date> <chr> <chr> <chr>
## 1 FF 2018-05-12 96.2 -2.1301 5.8072
## 2 FC 2018-05-12 88.8 -2.1697 5.8660
## 3 FC 2018-05-12 88.7 -2.2286 5.8761
## 4 FC 2018-05-12 88.4 -2.3845 5.7680
## 5 FF 2018-05-12 95.4 -2.2892 5.7659
## 6 FF 2018-05-12 94.9 -2.2536 5.8817
## # ... with 82 more variables: player_name <chr>, batter <int>,
## # pitcher <int>, events <chr>, description <chr>, spin_dir <chr>,
## # spin_rate_deprecated <chr>, break_angle_deprecated <chr>,
## # break_length_deprecated <chr>, zone <chr>, des <chr>, game_type <chr>,
## # stand <chr>, p_throws <chr>, home_team <chr>, away_team <chr>,
## # type <chr>, hit_location <chr>, bb_type <chr>, balls <int>,
## # strikes <int>, game_year <int>, pfx_x <chr>, pfx_z <chr>,
## # plate_x <chr>, plate_z <chr>, on_3b <chr>, on_2b <chr>, on_1b <chr>,
## # outs_when_up <int>, inning <int>, inning_topbot <chr>, hc_x <chr>,
## # hc_y <chr>, tfs_deprecated <chr>, tfs_zulu_deprecated <chr>,
## # pos2_person_id <chr>, umpire <chr>, sv_id <chr>, vx0 <chr>, vy0 <chr>,
## # vz0 <chr>, ax <chr>, ay <chr>, az <chr>, sz_top <chr>, sz_bot <chr>,
## # hit_distance_sc <chr>, launch_speed <chr>, launch_angle <chr>,
## # effective_speed <chr>, release_spin_rate <chr>,
## # release_extension <chr>, game_pk <int>, pos1_person_id <chr>,
## # pos2_person_id_1 <chr>, pos3_person_id <chr>, pos4_person_id <chr>,
## # pos5_person_id <chr>, pos6_person_id <chr>, pos7_person_id <chr>,
## # pos8_person_id <chr>, pos9_person_id <chr>, release_pos_y <chr>,
## # estimated_ba_using_speedangle <chr>,
## # estimated_woba_using_speedangle <chr>, woba_value <chr>,
## # woba_denom <chr>, babip_value <chr>, iso_value <chr>,
## # launch_speed_angle <chr>, at_bat_number <int>, pitch_number <int>,
## # pitch_name <chr>, home_score <int>, away_score <int>, bat_score <int>,
## # fld_score <int>, post_away_score <int>, post_home_score <int>,
## # post_bat_score <int>, post_fld_score <int>Create ID Tables
If we look at the pitching and batting data we will notice that there are fields with ids that are not translated to names. So we will use the batting and pitching data to generate ids for each person. Since the player_name is the name of the batter/pitcher in their respective data sets we should be able to get all the ids by extracting the player_name with the batter field for the batting table and the player_name field with the pitcher field for the pitching data.
batter_id_table <- final_batting_data_table %>%
group_by(player_name, batter) %>%
select(player_name, batter) %>%
distinct
pitcher_id_table <- final_pitching_data_table %>%
group_by(player_name, pitcher) %>%
select(player_name, pitcher) %>%
distinct
head(batter_id_table)## # A tibble: 6 x 2
## # Groups: player_name, batter [6]
## player_name batter
## <chr> <int>
## 1 Hanley Ramirez 434670
## 2 Andrew Benintendi 643217
## 3 Mookie Betts 605141
## 4 Christian Vazquez 543877
## 5 Jackie Bradley Jr. 598265
## 6 Eduardo Nunez 456488head(pitcher_id_table)## # A tibble: 6 x 2
## # Groups: player_name, pitcher [6]
## player_name pitcher
## <chr> <int>
## 1 Craig Kimbrel 518886
## 2 Joe Kelly 523260
## 3 Hector Velazquez 584171
## 4 Carson Smith 605476
## 5 David Price 456034
## 6 Brian Johnson 598271Get Mookie Betts Data
Mookie Betts is having a standout season. I would like to focus on data regarding Mookie’s at bats.
mookie_data <- final_batting_data_table %>%
filter(player_name=="Mookie Betts")
head(mookie_data)## # A tibble: 6 x 87
## pitch_type game_date release_speed release_pos_x release_pos_z
## <chr> <date> <chr> <chr> <chr>
## 1 FF 2018-05-12 94.9 -2.2536 5.8817
## 2 FF 2018-05-12 95.8 -2.2788 5.8093
## 3 FT 2018-05-12 97.3 -2.4584 5.7709
## 4 FF 2018-05-12 96.1 -2.2799 5.7348
## 5 FF 2018-05-12 95.9 -1.9824 5.7779
## 6 FF 2018-05-12 95.7 -2.3119 5.7480
## # ... with 82 more variables: player_name <chr>, batter <int>,
## # pitcher <int>, events <chr>, description <chr>, spin_dir <chr>,
## # spin_rate_deprecated <chr>, break_angle_deprecated <chr>,
## # break_length_deprecated <chr>, zone <chr>, des <chr>, game_type <chr>,
## # stand <chr>, p_throws <chr>, home_team <chr>, away_team <chr>,
## # type <chr>, hit_location <chr>, bb_type <chr>, balls <int>,
## # strikes <int>, game_year <int>, pfx_x <chr>, pfx_z <chr>,
## # plate_x <chr>, plate_z <chr>, on_3b <chr>, on_2b <chr>, on_1b <chr>,
## # outs_when_up <int>, inning <int>, inning_topbot <chr>, hc_x <chr>,
## # hc_y <chr>, tfs_deprecated <chr>, tfs_zulu_deprecated <chr>,
## # pos2_person_id <chr>, umpire <chr>, sv_id <chr>, vx0 <chr>, vy0 <chr>,
## # vz0 <chr>, ax <chr>, ay <chr>, az <chr>, sz_top <chr>, sz_bot <chr>,
## # hit_distance_sc <chr>, launch_speed <chr>, launch_angle <chr>,
## # effective_speed <chr>, release_spin_rate <chr>,
## # release_extension <chr>, game_pk <int>, pos1_person_id <chr>,
## # pos2_person_id_1 <chr>, pos3_person_id <chr>, pos4_person_id <chr>,
## # pos5_person_id <chr>, pos6_person_id <chr>, pos7_person_id <chr>,
## # pos8_person_id <chr>, pos9_person_id <chr>, release_pos_y <chr>,
## # estimated_ba_using_speedangle <chr>,
## # estimated_woba_using_speedangle <chr>, woba_value <chr>,
## # woba_denom <chr>, babip_value <chr>, iso_value <chr>,
## # launch_speed_angle <chr>, at_bat_number <int>, pitch_number <int>,
## # pitch_name <chr>, home_score <int>, away_score <int>, bat_score <int>,
## # fld_score <int>, post_away_score <int>, post_home_score <int>,
## # post_bat_score <int>, post_fld_score <int>Events
mookie_data <- mookie_data %>%
mutate(rbi=post_bat_score-bat_score)
mookie_data %>%
group_by(events) %>%
summarize(count=n())## # A tibble: 15 x 2
## events count
## <chr> <int>
## 1 double 15
## 2 field_error 2
## 3 field_out 61
## 4 force_out 3
## 5 grounded_into_double_play 2
## 6 hit_by_pitch 3
## 7 home_run 13
## 8 intent_walk 2
## 9 null 463
## 10 other_out 1
## 11 sac_fly 2
## 12 single 18
## 13 strikeout 17
## 14 triple 1
## 15 walk 15Lets look at the at bats but grouped based on pitch type.
mookie_data %>%
group_by(events, pitch_name) %>%
summarize(count=n())## # A tibble: 64 x 3
## # Groups: events [?]
## events pitch_name count
## <chr> <chr> <int>
## 1 double 2-Seam Fastball 1
## 2 double 4-Seam Fastball 6
## 3 double Changeup 3
## 4 double Curveball 1
## 5 double Slider 4
## 6 field_error Changeup 1
## 7 field_error Slider 1
## 8 field_out 2-Seam Fastball 8
## 9 field_out 4-Seam Fastball 27
## 10 field_out Changeup 10
## # ... with 54 more rowsSome raw data links to column names https://fastballs.wordpress.com/category/pitchfx-glossary/
pfx_x: the horizontal movement, in inches, of the pitch between the release point and home plate, as compared to a theoretical pitch thrown at the same speed with no spin-induced movement. This parameter is measured at y=40 feet regardless of the y0 value.
pfx_z: the vertical movement, in inches, of the pitch between the release point and home plate, as compared to a theoretical pitch thrown at the same speed with no spin-induced movement. This parameter is measured at y=40 feet regardless of the y0 value.
plate_x: the left/right distance, in feet, of the pitch from the middle of the plate as it crossed home plate. The PITCHf/x coordinate system is oriented to the catcher’s/umpire’s perspective, with distances to the right being positive and to the left being negative.
plate_z: the height of the pitch in feet as it crossed the front of home plate.
mookie_bat_data <- mookie_data %>%
select(events, pitch_name, release_speed, pitcher, bb_type, pfx_x, pfx_z, plate_x, plate_z, launch_angle, launch_speed, hit_distance_sc, sz_top, sz_bot)
mookie_bat_data %>%
filter(events=="home_run") %>%
head()## # A tibble: 6 x 14
## events pitch_name release_speed pitcher bb_type pfx_x pfx_z plate_x
## <chr> <chr> <chr> <int> <chr> <chr> <chr> <chr>
## 1 home_r~ 2-Seam Fast~ 86.0 112526 fly_ball -1.2~ 0.76~ -0.2037
## 2 home_r~ 4-Seam Fast~ 92.9 501985 line_dri~ 0.85~ 1.60~ 0.3792
## 3 home_r~ 2-Seam Fast~ 92.9 518633 fly_ball 0.90~ 1.30~ 0.7083
## 4 home_r~ Slider 86.8 518633 fly_ball -0.2~ 0.13~ -0.0138
## 5 home_r~ Changeup 83.9 518633 fly_ball 1.12~ 1.21~ -0.3759
## 6 home_r~ 4-Seam Fast~ 91.6 592130 fly_ball -1.1~ 1.32~ 0.6080
## # ... with 6 more variables: plate_z <chr>, launch_angle <chr>,
## # launch_speed <chr>, hit_distance_sc <chr>, sz_top <chr>, sz_bot <chr>Before we can analyze and graph this we need to clean it up.
mookie_bat_data <- mookie_bat_data %>%
type_convert(cols(release_speed=col_double(), pfx_x=col_double(), pfx_z=col_double(), plate_x=col_double(), plate_z=col_double(), launch_angle=col_double(), launch_speed=col_double(), hit_distance_sc=col_integer(), sz_top=col_double(), sz_bot=col_double()))
mookie_bat_data$pitch_name <- as.factor(mookie_bat_data$pitch_name)
mookie_bat_data$events <- as.factor(mookie_bat_data$events)
mookie_bat_data$bb_type <- as.factor(mookie_bat_data$bb_type)
head(mookie_bat_data)## # A tibble: 6 x 14
## events pitch_name release_speed pitcher bb_type pfx_x pfx_z plate_x
## <fct> <fct> <dbl> <int> <fct> <dbl> <dbl> <dbl>
## 1 double 4-Seam Fastb~ 94.9 572193 line_dr~ -0.546 1.23 -0.684
## 2 null 4-Seam Fastb~ 95.8 572193 null -0.669 1.23 0.837
## 3 null 2-Seam Fastb~ 97.3 572193 null -1.31 1.14 1.24
## 4 null 4-Seam Fastb~ 96.1 572193 null -0.361 1.47 -0.718
## 5 null 4-Seam Fastb~ 95.9 572193 null -0.432 1.52 1.73
## 6 null 4-Seam Fastb~ 95.7 572193 null -0.711 1.30 0.263
## # ... with 6 more variables: plate_z <dbl>, launch_angle <dbl>,
## # launch_speed <dbl>, hit_distance_sc <int>, sz_top <dbl>, sz_bot <dbl>strike_zone_top <- mean(mookie_bat_data$sz_top, na.rm=TRUE)
strike_zone_bot <- mean(mookie_bat_data$sz_bot, na.rm=TRUE)
strike_zone_front <- 17/2.0/12
graphed <- ggplot() + geom_rect(aes(xmin=-strike_zone_front, xmax=strike_zone_front, ymin=strike_zone_bot, ymax=strike_zone_top), color="black", fill=NA) + geom_point(data=mookie_bat_data, aes(x=plate_x, y=plate_z, text=sprintf("events: %s<br>Pitch Name: %s<br>Release Speed: %s<br>Launch Angle: %s", events, pitch_name, release_speed, launch_angle))) +
scale_x_continuous(name="Catchers View | left/right distance in feet of center of plate", limits=c(-4, 4), breaks=seq(from=-4, to=4, by=2)) + scale_y_continuous(name="Height above plate In Feet", limits=c(-2, 6), breaks=seq(from=-2, to=6, by=1))
ggplotly(graphed) action <- mookie_bat_data %>%
filter(events!="null")
graphed <- ggplot() + geom_rect(aes(xmin=-strike_zone_front, xmax=strike_zone_front, ymin=strike_zone_bot, ymax=strike_zone_top), color="black", fill=NA) + geom_point(data=action, aes(x=plate_x, y=plate_z, text=sprintf("events: %s<br>Pitch Name: %s<br>Release Speed: %s<br>Launch Angle: %s", events, pitch_name, release_speed, launch_angle), color=events)) +
scale_x_continuous(name="Catchers View | left/right distance in feet of center of plate", limits=c(-4, 4), breaks=seq(from=-4, to=4, by=2)) + scale_y_continuous(name="Height above plate In Feet", limits=c(-2, 6), breaks=seq(from=-2, to=6, by=1))
ggplotly(graphed) action <- mookie_bat_data %>%
filter(events == "home_run")
graphed <- ggplot() + geom_rect(aes(xmin=-strike_zone_front, xmax=strike_zone_front, ymin=strike_zone_bot, ymax=strike_zone_top), color="black", fill=NA) + geom_point(data=action, aes(x=plate_x, y=plate_z, text=sprintf("events: %s<br>Pitch Name: %s<br>Release Speed: %s<br>Launch Angle: %s", events, pitch_name, release_speed, launch_angle), color=pitch_name)) +
scale_x_continuous(name="Catchers View | left/right distance in feet of center of plate", limits=c(-4, 4), breaks=seq(from=-4, to=4, by=2)) + scale_y_continuous(name="Height above plate In Feet", limits=c(-2, 6), breaks=seq(from=-2, to=6, by=1))
ggplotly(graphed) Lets try to predict the outcome of a pitch.
set.seed(1234)
mookie_bat_data_converted <- mookie_bat_data %>%
filter(events!="null") %>%
filter(events!="intent_walk") %>%
filter(!is.na(plate_x)) %>%
select(events, pitch_name, release_speed, plate_x, plate_z)
mookie_bat_data_converted ## # A tibble: 151 x 5
## events pitch_name release_speed plate_x plate_z
## <fct> <fct> <dbl> <dbl> <dbl>
## 1 double 4-Seam Fastball 94.9 -0.684 3.03
## 2 single 2-Seam Fastball 92.7 -0.661 2.68
## 3 field_out Changeup 74.9 -0.0167 3.39
## 4 double Changeup 77.8 0.0608 1.58
## 5 field_out 4-Seam Fastball 89.8 0.404 2.85
## 6 strikeout Slider 84.3 1.34 1.21
## 7 field_error Slider 86.0 0.646 3.05
## 8 field_out Cutter 85.5 0.471 2.67
## 9 field_out 2-Seam Fastball 93.8 -0.765 2.74
## 10 strikeout Curveball 78.8 0.990 1.92
## # ... with 141 more rowsmookie_tree <- tree(events~pitch_name+plate_x+plate_z+release_speed, data=mookie_bat_data_converted)
# We will save the plot as a png to deal with overlapping graphics
png("image.png", width = 1200, height = 700)
plot(mookie_tree)
text(mookie_tree, pretty=1, cex=1.1)
dev.off()## png
## 2img <- readPNG('image.png')
grid::grid.raster(img)
This tree shows us the what the likelihood outcome of a pitch is based on location, type, and speed.
Lets try and use a Random Forest
mookie_bat_data_converted$pitch_name = as.numeric(mookie_bat_data_converted$pitch_name)
mookie_bat_data_converted <- droplevels(mookie_bat_data_converted)
train_set <- mookie_bat_data_converted %>%
group_by(events) %>%
sample_frac(.6) %>%
ungroup()
test_set <- mookie_bat_data_converted %>%
anti_join(train_set)
train_set## # A tibble: 91 x 5
## events pitch_name release_speed plate_x plate_z
## <fct> <dbl> <dbl> <dbl> <dbl>
## 1 double 3.00 77.8 0.0608 1.58
## 2 double 2.00 90.4 -0.630 1.69
## 3 double 4.00 77.3 -0.00490 2.43
## 4 double 2.00 92.9 -0.208 1.72
## 5 double 8.00 84.9 0.767 1.64
## 6 double 3.00 88.4 0.686 2.01
## 7 double 2.00 94.9 -0.684 3.03
## 8 double 8.00 82.1 -0.0844 2.40
## 9 double 3.00 82.3 -0.0869 1.05
## 10 field_error 3.00 83.0 0.0677 1.62
## # ... with 81 more rowstest_set## # A tibble: 60 x 5
## events pitch_name release_speed plate_x plate_z
## <fct> <dbl> <dbl> <dbl> <dbl>
## 1 field_out 3.00 74.9 -0.0167 3.39
## 2 field_error 8.00 86.0 0.646 3.05
## 3 field_out 5.00 85.5 0.471 2.67
## 4 field_out 1.00 93.8 -0.765 2.74
## 5 walk 2.00 95.1 -0.0117 3.85
## 6 field_out 2.00 98.0 0.260 2.08
## 7 single 3.00 83.1 0.609 2.16
## 8 single 2.00 91.5 -0.0774 2.62
## 9 double 8.00 80.9 0.701 2.29
## 10 field_out 2.00 96.7 -0.216 3.23
## # ... with 50 more rowsmookie_rf <- randomForest(events~pitch_name+plate_x+plate_z, data=train_set)
test_predictions <- predict(mookie_rf, newdata=test_set)
cm <- table(pred=test_predictions, observed=test_set$events)
confusionMatrix(cm)## Confusion Matrix and Statistics
##
## observed
## pred double field_error field_out force_out
## double 0 0 0 0
## field_error 0 0 0 0
## field_out 5 0 22 1
## force_out 0 0 0 0
## grounded_into_double_play 0 0 0 0
## hit_by_pitch 0 0 0 0
## home_run 0 0 0 0
## other_out 0 0 0 0
## sac_fly 0 0 0 0
## single 0 0 1 0
## strikeout 1 1 0 0
## triple 0 0 0 0
## walk 0 0 1 0
## observed
## pred grounded_into_double_play hit_by_pitch
## double 0 0
## field_error 0 0
## field_out 1 0
## force_out 0 0
## grounded_into_double_play 0 0
## hit_by_pitch 0 0
## home_run 0 0
## other_out 0 0
## sac_fly 0 0
## single 0 0
## strikeout 0 0
## triple 0 0
## walk 0 1
## observed
## pred home_run other_out sac_fly single strikeout
## double 1 0 0 1 0
## field_error 0 0 0 0 0
## field_out 3 0 1 6 4
## force_out 0 0 0 0 0
## grounded_into_double_play 0 0 0 0 0
## hit_by_pitch 0 0 0 0 0
## home_run 0 0 0 0 0
## other_out 0 0 0 0 0
## sac_fly 0 0 0 0 0
## single 1 0 0 0 0
## strikeout 0 0 0 0 3
## triple 0 0 0 0 0
## walk 0 0 0 0 0
## observed
## pred triple walk
## double 0 1
## field_error 0 0
## field_out 0 2
## force_out 0 0
## grounded_into_double_play 0 0
## hit_by_pitch 0 0
## home_run 0 0
## other_out 0 0
## sac_fly 0 0
## single 0 0
## strikeout 0 0
## triple 0 0
## walk 0 3
##
## Overall Statistics
##
## Accuracy : 0.4667
## 95% CI : (0.3367, 0.6)
## No Information Rate : 0.4
## P-Value [Acc > NIR] : 0.1779
##
## Kappa : 0.2076
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: double Class: field_error Class: field_out
## Sensitivity 0.0000 0.00000 0.9167
## Specificity 0.9444 1.00000 0.3611
## Pos Pred Value 0.0000 NaN 0.4889
## Neg Pred Value 0.8947 0.98333 0.8667
## Prevalence 0.1000 0.01667 0.4000
## Detection Rate 0.0000 0.00000 0.3667
## Detection Prevalence 0.0500 0.00000 0.7500
## Balanced Accuracy 0.4722 0.50000 0.6389
## Class: force_out Class: grounded_into_double_play
## Sensitivity 0.00000 0.00000
## Specificity 1.00000 1.00000
## Pos Pred Value NaN NaN
## Neg Pred Value 0.98333 0.98333
## Prevalence 0.01667 0.01667
## Detection Rate 0.00000 0.00000
## Detection Prevalence 0.00000 0.00000
## Balanced Accuracy 0.50000 0.50000
## Class: hit_by_pitch Class: home_run Class: other_out
## Sensitivity 0.00000 0.00000 NA
## Specificity 1.00000 1.00000 1
## Pos Pred Value NaN NaN NA
## Neg Pred Value 0.98333 0.91667 NA
## Prevalence 0.01667 0.08333 0
## Detection Rate 0.00000 0.00000 0
## Detection Prevalence 0.00000 0.00000 0
## Balanced Accuracy 0.50000 0.50000 NA
## Class: sac_fly Class: single Class: strikeout
## Sensitivity 0.00000 0.00000 0.42857
## Specificity 1.00000 0.96226 0.96226
## Pos Pred Value NaN 0.00000 0.60000
## Neg Pred Value 0.98333 0.87931 0.92727
## Prevalence 0.01667 0.11667 0.11667
## Detection Rate 0.00000 0.00000 0.05000
## Detection Prevalence 0.00000 0.03333 0.08333
## Balanced Accuracy 0.50000 0.48113 0.69542
## Class: triple Class: walk
## Sensitivity NA 0.50000
## Specificity 1 0.96296
## Pos Pred Value NA 0.60000
## Neg Pred Value NA 0.94545
## Prevalence 0 0.10000
## Detection Rate 0 0.05000
## Detection Prevalence 0 0.08333
## Balanced Accuracy NA 0.7314846% accurate is really bad. So there might not be enough data on mookie to predict properly. What if we try on all the red sox batters with as many predictors as we can get. We forgot that a 3-0 count is never gonna result in a strikeout whereas a 0-2 will never get a walk, and if no one is one base there can’t be a doulbe play. So we add as many predictors as we can.
# We filter out na's and select our columns.
final_batting_data_table_converted <- final_batting_data_table %>%
mutate(release_speed=replace(release_speed, release_speed=="null", 0))%>%
mutate(pfx_x=replace(pfx_x, pfx_x=="null", 0))%>%
mutate(pfx_z=replace(pfx_z, pfx_z=="null", 0))%>%
mutate(plate_x=replace(plate_x, plate_x=="null", 0))%>%
mutate(plate_z=replace(plate_z, plate_z=="null", 0))%>%
mutate(on_1b=replace(on_1b, on_1b=="null", 0))%>%
mutate(on_2b=replace(on_2b, on_2b=="null", 0))%>%
mutate(on_3b=replace(on_3b, on_3b=="null", 0))%>%
mutate(hit_location=replace(hit_location, hit_location=="null", 0))%>%
mutate(hit_distance_sc=replace(hit_distance_sc, hit_distance_sc=="null", 0))%>%
mutate(launch_angle=replace(launch_angle, launch_angle=="null", 0))%>%
mutate(launch_speed=replace(launch_speed, launch_speed=="null", 0))%>%
type_convert(cols(release_speed=col_double(), pfx_x=col_double(), pfx_z=col_double(), plate_x=col_double(), plate_z=col_double(), on_1b=col_double(), on_2b=col_double(), on_3b=col_double(), hit_distance_sc=col_double(), launch_angle=col_double(), launch_speed=col_double(), hit_location=col_double())) %>%
filter(events!="intent_walk") %>%
filter(events!="null") %>%
filter(!is.na(plate_x)) %>%
filter(!is.na(plate_z)) %>%
filter(!is.na(pitch_name)) %>%
filter(!is.na(release_speed)) %>%
mutate(rbi=post_bat_score-bat_score) %>%
select(events, pitch_name, release_speed, plate_x, plate_z, pfx_x, pfx_z, batter, pitcher, balls, strikes, outs_when_up, on_1b, on_2b, on_3b, hit_location, pfx_x, pfx_z, hit_distance_sc, launch_speed, launch_angle, pitch_number, rbi)final_batting_data_table_converted$pitch_name <- as.factor(final_batting_data_table_converted$pitch_name)
final_batting_data_table_converted$events <- as.factor(final_batting_data_table_converted$events)
final_batting_data_table_converted$on_1b[is.na(final_batting_data_table_converted$on_1b)] <- 0
final_batting_data_table_converted$on_2b[is.na(final_batting_data_table_converted$on_2b)] <- 0
final_batting_data_table_converted$on_3b[is.na(final_batting_data_table_converted$on_3b)] <- 0
# We convert the pitch name to numeric
final_batting_data_table_converted$pitch_name = as.numeric(final_batting_data_table_converted$pitch_name)
# We drop unused levels
final_batting_data_table_converted <- droplevels(final_batting_data_table_converted)
# Use 60/40 split for train data / test data
train_set <- final_batting_data_table_converted %>%
group_by(events) %>%
sample_frac(.8) %>%
ungroup()
test_set <- final_batting_data_table_converted %>%
anti_join(train_set)## Joining, by = c("events", "pitch_name", "release_speed", "plate_x", "plate_z", "pfx_x", "pfx_z", "batter", "pitcher", "balls", "strikes", "outs_when_up", "on_1b", "on_2b", "on_3b", "hit_location", "hit_distance_sc", "launch_speed", "launch_angle", "pitch_number", "rbi")train_set## # A tibble: 1,238 x 21
## events pitch_name release_speed plate_x plate_z pfx_x pfx_z batter
## <fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <int>
## 1 caught_s~ 2.00 93.1 -0.364 1.73 -0.716 1.72 643217
## 2 caught_s~ 8.00 93.9 1.09 2.37 -0.926 1.22 519048
## 3 double 2.00 91.2 -0.648 2.89 -0.769 1.48 543877
## 4 double 1.00 92.7 -0.508 1.46 -1.36 0.801 643217
## 5 double 1.00 92.7 0.0661 1.29 -1.36 0.924 646240
## 6 double 9.00 81.6 0.513 2.13 0.830 -0.319 646240
## 7 double 2.00 90.9 0.515 3.12 -0.764 1.76 519048
## 8 double 5.00 88.5 -0.162 1.79 0.196 0.565 593428
## 9 double 4.00 81.3 -0.611 3.06 -0.562 -0.928 543877
## 10 double 3.00 81.1 0.249 2.40 1.64 0.392 593428
## # ... with 1,228 more rows, and 13 more variables: pitcher <int>,
## # balls <int>, strikes <int>, outs_when_up <int>, on_1b <dbl>,
## # on_2b <dbl>, on_3b <dbl>, hit_location <dbl>, hit_distance_sc <dbl>,
## # launch_speed <dbl>, launch_angle <dbl>, pitch_number <int>, rbi <int>test_set## # A tibble: 309 x 21
## events pitch_name release_speed plate_x plate_z pfx_x pfx_z batter
## <fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <int>
## 1 field_out 3.00 75.6 -0.880 2.11 -1.21 1.59 643217
## 2 field_out 2.00 89.2 1.32 3.13 -0.258 1.92 456488
## 3 strikeout 3.00 75.9 -0.884 1.60 -1.02 1.54 646240
## 4 field_out 2.00 89.8 0.404 2.85 0.0873 1.87 605141
## 5 field_out 9.00 83.9 1.07 2.45 0.341 0.624 643217
## 6 strikeout 2.00 91.7 -0.929 2.49 -0.331 1.61 571788
## 7 strikeout 2.00 93.1 0.0445 3.22 -0.563 1.51 643217
## 8 field_out 5.00 85.5 0.471 2.67 0.189 0.493 605141
## 9 strikeout 3.00 82.9 -0.506 0.507 -1.51 0.526 506702
## 10 field_out 3.00 86.6 -0.723 1.41 -1.63 0.368 593428
## # ... with 299 more rows, and 13 more variables: pitcher <int>,
## # balls <int>, strikes <int>, outs_when_up <int>, on_1b <dbl>,
## # on_2b <dbl>, on_3b <dbl>, hit_location <dbl>, hit_distance_sc <dbl>,
## # launch_speed <dbl>, launch_angle <dbl>, pitch_number <int>, rbi <int>final_rf <- randomForest(events~batter+pitcher+pitch_name+plate_x+plate_z+release_speed+balls+strikes+outs_when_up+on_1b+on_2b+on_3b+hit_location+pfx_x+pfx_z+hit_distance_sc+launch_speed+launch_angle+pitch_number+rbi, data=train_set)
test_predictions <- predict(final_rf, newdata=test_set)
cm <- table(pred=test_predictions, observed=test_set$events)
confusionMatrix(cm)## Confusion Matrix and Statistics
##
## observed
## pred caught_stealing_2b double double_play
## caught_stealing_2b 0 0 0
## double 0 9 0
## double_play 0 0 0
## field_error 0 0 0
## field_out 0 1 1
## fielders_choice 0 0 0
## fielders_choice_out 0 0 0
## force_out 0 0 0
## grounded_into_double_play 0 0 0
## hit_by_pitch 0 0 0
## home_run 0 2 0
## other_out 0 0 0
## pickoff_2b 0 0 0
## sac_bunt 0 0 0
## sac_fly 0 0 0
## single 0 7 0
## strikeout 0 0 0
## strikeout_double_play 0 0 0
## triple 0 0 0
## walk 0 0 0
## observed
## pred field_error field_out fielders_choice
## caught_stealing_2b 0 0 0
## double 0 0 0
## double_play 0 0 0
## field_error 0 0 0
## field_out 4 126 0
## fielders_choice 0 0 0
## fielders_choice_out 0 0 0
## force_out 0 0 0
## grounded_into_double_play 0 0 0
## hit_by_pitch 0 0 0
## home_run 0 0 0
## other_out 0 0 0
## pickoff_2b 0 0 0
## sac_bunt 0 0 0
## sac_fly 0 0 0
## single 0 0 0
## strikeout 0 0 0
## strikeout_double_play 0 0 0
## triple 0 0 0
## walk 0 0 0
## observed
## pred fielders_choice_out force_out
## caught_stealing_2b 0 0
## double 0 0
## double_play 0 0
## field_error 0 0
## field_out 1 6
## fielders_choice 0 0
## fielders_choice_out 0 0
## force_out 0 0
## grounded_into_double_play 0 0
## hit_by_pitch 0 0
## home_run 0 0
## other_out 0 0
## pickoff_2b 0 0
## sac_bunt 0 0
## sac_fly 0 0
## single 0 0
## strikeout 0 0
## strikeout_double_play 0 0
## triple 0 0
## walk 0 0
## observed
## pred grounded_into_double_play hit_by_pitch
## caught_stealing_2b 0 0
## double 0 0
## double_play 0 0
## field_error 0 0
## field_out 4 0
## fielders_choice 0 0
## fielders_choice_out 0 0
## force_out 0 0
## grounded_into_double_play 1 0
## hit_by_pitch 0 3
## home_run 0 0
## other_out 0 0
## pickoff_2b 0 0
## sac_bunt 0 0
## sac_fly 0 0
## single 0 0
## strikeout 0 0
## strikeout_double_play 0 0
## triple 0 0
## walk 0 0
## observed
## pred home_run other_out pickoff_2b sac_bunt sac_fly
## caught_stealing_2b 0 0 0 0 0
## double 2 0 0 0 0
## double_play 0 0 0 0 0
## field_error 0 0 0 0 0
## field_out 0 0 0 0 3
## fielders_choice 0 0 0 0 0
## fielders_choice_out 0 0 0 0 0
## force_out 0 0 0 0 0
## grounded_into_double_play 0 0 0 0 0
## hit_by_pitch 0 0 0 0 0
## home_run 9 0 0 0 0
## other_out 0 0 0 0 0
## pickoff_2b 0 0 0 0 0
## sac_bunt 0 0 0 0 0
## sac_fly 0 0 0 0 0
## single 0 0 0 0 0
## strikeout 0 0 0 0 0
## strikeout_double_play 0 0 0 0 0
## triple 0 0 0 0 0
## walk 0 0 0 0 0
## observed
## pred single strikeout strikeout_double_play triple
## caught_stealing_2b 0 0 0 0
## double 4 0 0 0
## double_play 0 0 0 0
## field_error 0 0 0 0
## field_out 1 1 0 0
## fielders_choice 0 0 0 0
## fielders_choice_out 0 0 0 0
## force_out 0 0 0 0
## grounded_into_double_play 0 0 0 0
## hit_by_pitch 0 0 0 0
## home_run 0 0 0 0
## other_out 0 0 0 0
## pickoff_2b 0 0 0 0
## sac_bunt 0 0 0 0
## sac_fly 0 0 0 0
## single 39 0 0 1
## strikeout 0 62 0 0
## strikeout_double_play 0 0 0 0
## triple 0 0 0 0
## walk 0 0 0 0
## observed
## pred walk
## caught_stealing_2b 0
## double 0
## double_play 0
## field_error 0
## field_out 0
## fielders_choice 0
## fielders_choice_out 0
## force_out 0
## grounded_into_double_play 0
## hit_by_pitch 0
## home_run 0
## other_out 0
## pickoff_2b 0
## sac_bunt 0
## sac_fly 0
## single 0
## strikeout 0
## strikeout_double_play 0
## triple 0
## walk 22
##
## Overall Statistics
##
## Accuracy : 0.877
## 95% CI : (0.8351, 0.9115)
## No Information Rate : 0.4078
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.8321
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: caught_stealing_2b Class: double
## Sensitivity NA 0.47368
## Specificity 1 0.97931
## Pos Pred Value NA 0.60000
## Neg Pred Value NA 0.96599
## Prevalence 0 0.06149
## Detection Rate 0 0.02913
## Detection Prevalence 0 0.04854
## Balanced Accuracy NA 0.72650
## Class: double_play Class: field_error
## Sensitivity 0.000000 0.00000
## Specificity 1.000000 1.00000
## Pos Pred Value NaN NaN
## Neg Pred Value 0.996764 0.98706
## Prevalence 0.003236 0.01294
## Detection Rate 0.000000 0.00000
## Detection Prevalence 0.000000 0.00000
## Balanced Accuracy 0.500000 0.50000
## Class: field_out Class: fielders_choice
## Sensitivity 1.0000 NA
## Specificity 0.8798 1
## Pos Pred Value 0.8514 NA
## Neg Pred Value 1.0000 NA
## Prevalence 0.4078 0
## Detection Rate 0.4078 0
## Detection Prevalence 0.4790 0
## Balanced Accuracy 0.9399 NA
## Class: fielders_choice_out Class: force_out
## Sensitivity 0.000000 0.00000
## Specificity 1.000000 1.00000
## Pos Pred Value NaN NaN
## Neg Pred Value 0.996764 0.98058
## Prevalence 0.003236 0.01942
## Detection Rate 0.000000 0.00000
## Detection Prevalence 0.000000 0.00000
## Balanced Accuracy 0.500000 0.50000
## Class: grounded_into_double_play Class: hit_by_pitch
## Sensitivity 0.200000 1.000000
## Specificity 1.000000 1.000000
## Pos Pred Value 1.000000 1.000000
## Neg Pred Value 0.987013 1.000000
## Prevalence 0.016181 0.009709
## Detection Rate 0.003236 0.009709
## Detection Prevalence 0.003236 0.009709
## Balanced Accuracy 0.600000 1.000000
## Class: home_run Class: other_out Class: pickoff_2b
## Sensitivity 0.81818 NA NA
## Specificity 0.99329 1 1
## Pos Pred Value 0.81818 NA NA
## Neg Pred Value 0.99329 NA NA
## Prevalence 0.03560 0 0
## Detection Rate 0.02913 0 0
## Detection Prevalence 0.03560 0 0
## Balanced Accuracy 0.90574 NA NA
## Class: sac_bunt Class: sac_fly Class: single
## Sensitivity NA 0.000000 0.8864
## Specificity 1 1.000000 0.9698
## Pos Pred Value NA NaN 0.8298
## Neg Pred Value NA 0.990291 0.9809
## Prevalence 0 0.009709 0.1424
## Detection Rate 0 0.000000 0.1262
## Detection Prevalence 0 0.000000 0.1521
## Balanced Accuracy NA 0.500000 0.9281
## Class: strikeout Class: strikeout_double_play
## Sensitivity 0.9841 NA
## Specificity 1.0000 1
## Pos Pred Value 1.0000 NA
## Neg Pred Value 0.9960 NA
## Prevalence 0.2039 0
## Detection Rate 0.2006 0
## Detection Prevalence 0.2006 0
## Balanced Accuracy 0.9921 NA
## Class: triple Class: walk
## Sensitivity 0.000000 1.0000
## Specificity 1.000000 1.0000
## Pos Pred Value NaN 1.0000
## Neg Pred Value 0.996764 1.0000
## Prevalence 0.003236 0.0712
## Detection Rate 0.000000 0.0712
## Detection Prevalence 0.000000 0.0712
## Balanced Accuracy 0.500000 1.000085% accuracy is a lot better. By adding more predictors we were able to drastically improve our accuracy.